; AVX1-NEXT: pushq %r13
; AVX1-NEXT: pushq %r12
; AVX1-NEXT: pushq %rbx
-; AVX1-NEXT: subq $16, %rsp
+; AVX1-NEXT: subq $24, %rsp
; AVX1-NEXT: vpmovzxbw {{.*#+}} xmm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
; AVX1-NEXT: vpmovzxbw {{.*#+}} xmm3 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
; AVX1-NEXT: vpmovzxbw {{.*#+}} xmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
; AVX1-NEXT: vmovq %xmm5, %rbp
; AVX1-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX1-NEXT: vpextrq $1, %xmm4, %rcx
-; AVX1-NEXT: vmovq %xmm4, %rsi
+; AVX1-NEXT: vpextrq $1, %xmm4, %rsi
+; AVX1-NEXT: vmovq %xmm4, %rcx
; AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
; AVX1-NEXT: vpmovzxwd {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
-; AVX1-NEXT: vpextrq $1, %xmm4, %r10
+; AVX1-NEXT: vpextrq $1, %xmm4, %r8
; AVX1-NEXT: vmovq %xmm4, %r11
; AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero
; AVX1-NEXT: vpmovzxwd {{.*#+}} xmm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
; AVX1-NEXT: vpextrq $1, %xmm4, %r15
-; AVX1-NEXT: vmovq %xmm4, %rdx
+; AVX1-NEXT: vmovq %xmm4, %rdi
; AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero
; AVX1-NEXT: vpextrq $1, %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
; AVX1-NEXT: vpmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm3 = xmm1[0],zero,xmm1[1],zero
; AVX1-NEXT: vpextrq $1, %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; AVX1-NEXT: vmovq %xmm3, %r9
+; AVX1-NEXT: vmovq %xmm3, %r10
; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
; AVX1-NEXT: vpmovzxwd {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
-; AVX1-NEXT: vpextrq $1, %xmm4, %r14
-; AVX1-NEXT: addq %rbx, %r14
-; AVX1-NEXT: vmovq %xmm4, %r8
-; AVX1-NEXT: addq %rbp, %r8
+; AVX1-NEXT: vpextrq $1, %xmm4, %rdx
+; AVX1-NEXT: addq %rbx, %rdx
+; AVX1-NEXT: vmovq %xmm4, %r9
+; AVX1-NEXT: addq %rbp, %r9
; AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero
-; AVX1-NEXT: vpextrq $1, %xmm3, %rdi
-; AVX1-NEXT: addq %rcx, %rdi
-; AVX1-NEXT: vmovq %xmm3, %rax
+; AVX1-NEXT: vpextrq $1, %xmm3, %rax
; AVX1-NEXT: addq %rsi, %rax
-; AVX1-NEXT: movq %rax, %rsi
+; AVX1-NEXT: movq %rax, %r14
+; AVX1-NEXT: vmovq %xmm3, %rbp
+; AVX1-NEXT: addq %rcx, %rbp
; AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
; AVX1-NEXT: vpmovzxwd {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT: vpextrq $1, %xmm3, %rax
-; AVX1-NEXT: addq %r10, %rax
-; AVX1-NEXT: movq %rax, %r10
+; AVX1-NEXT: vpextrq $1, %xmm3, %rsi
+; AVX1-NEXT: addq %r8, %rsi
; AVX1-NEXT: vmovq %xmm3, %rax
; AVX1-NEXT: addq %r11, %rax
; AVX1-NEXT: movq %rax, %r11
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
; AVX1-NEXT: vpextrq $1, %xmm2, %rax
; AVX1-NEXT: addq %r13, %rax
-; AVX1-NEXT: movq %rax, %rbx
+; AVX1-NEXT: movq %rax, %rcx
; AVX1-NEXT: vmovq %xmm2, %rax
; AVX1-NEXT: addq %r12, %rax
-; AVX1-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: movq %rax, %r8
; AVX1-NEXT: vpmovzxwd {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
; AVX1-NEXT: vpextrq $1, %xmm3, %rax
; AVX1-NEXT: addq %r15, %rax
-; AVX1-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: movq %rax, %rbx
; AVX1-NEXT: vmovq %xmm3, %rax
-; AVX1-NEXT: addq %rdx, %rax
+; AVX1-NEXT: addq %rdi, %rax
; AVX1-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
; AVX1-NEXT: vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero
-; AVX1-NEXT: vpextrq $1, %xmm2, %rbp
-; AVX1-NEXT: addq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Folded Reload
-; AVX1-NEXT: vmovq %xmm2, %r15
-; AVX1-NEXT: addq %r9, %r15
+; AVX1-NEXT: vpextrq $1, %xmm2, %rax
+; AVX1-NEXT: addq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Folded Reload
+; AVX1-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: vmovq %xmm2, %r12
+; AVX1-NEXT: addq %r10, %r12
; AVX1-NEXT: vpextrq $1, %xmm1, %rax
; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
; AVX1-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX1-NEXT: vpextrq $1, %xmm0, %r9
-; AVX1-NEXT: addq %rax, %r9
-; AVX1-NEXT: vmovq %xmm1, %rcx
-; AVX1-NEXT: vmovq %xmm0, %rdx
-; AVX1-NEXT: addq %rcx, %rdx
-; AVX1-NEXT: addq $-1, %r14
-; AVX1-NEXT: movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: vpextrq $1, %xmm0, %r10
+; AVX1-NEXT: addq %rax, %r10
+; AVX1-NEXT: vmovq %xmm1, %rax
+; AVX1-NEXT: vmovq %xmm0, %rdi
+; AVX1-NEXT: addq %rax, %rdi
+; AVX1-NEXT: addq $-1, %rdx
+; AVX1-NEXT: movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX1-NEXT: movl $0, %eax
; AVX1-NEXT: adcq $-1, %rax
; AVX1-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX1-NEXT: addq $-1, %r8
-; AVX1-NEXT: movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: addq $-1, %r9
+; AVX1-NEXT: movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX1-NEXT: movl $0, %eax
; AVX1-NEXT: adcq $-1, %rax
-; AVX1-NEXT: movq %rax, (%rsp) # 8-byte Spill
-; AVX1-NEXT: addq $-1, %rdi
-; AVX1-NEXT: movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: addq $-1, %r14
+; AVX1-NEXT: movq %r14, (%rsp) # 8-byte Spill
; AVX1-NEXT: movl $0, %eax
; AVX1-NEXT: adcq $-1, %rax
; AVX1-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX1-NEXT: addq $-1, %rsi
-; AVX1-NEXT: movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: addq $-1, %rbp
+; AVX1-NEXT: movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX1-NEXT: movl $0, %eax
; AVX1-NEXT: adcq $-1, %rax
; AVX1-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX1-NEXT: addq $-1, %r10
-; AVX1-NEXT: movq %r10, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: addq $-1, %rsi
+; AVX1-NEXT: movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX1-NEXT: movl $0, %eax
; AVX1-NEXT: adcq $-1, %rax
; AVX1-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX1-NEXT: movl $0, %eax
; AVX1-NEXT: adcq $-1, %rax
; AVX1-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: addq $-1, %rcx
+; AVX1-NEXT: movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: movl $0, %ebp
+; AVX1-NEXT: adcq $-1, %rbp
+; AVX1-NEXT: addq $-1, %r8
+; AVX1-NEXT: movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: movl $0, %r15d
+; AVX1-NEXT: adcq $-1, %r15
; AVX1-NEXT: addq $-1, %rbx
; AVX1-NEXT: movq %rbx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX1-NEXT: movl $0, %eax
; AVX1-NEXT: adcq $-1, %rax
; AVX1-NEXT: movq %rax, %rsi
; AVX1-NEXT: addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; AVX1-NEXT: movl $0, %r12d
-; AVX1-NEXT: adcq $-1, %r12
-; AVX1-NEXT: addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
; AVX1-NEXT: movl $0, %r13d
; AVX1-NEXT: adcq $-1, %r13
; AVX1-NEXT: addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
; AVX1-NEXT: movl $0, %r14d
; AVX1-NEXT: adcq $-1, %r14
-; AVX1-NEXT: addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; AVX1-NEXT: movl $0, %ebx
-; AVX1-NEXT: adcq $-1, %rbx
-; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; AVX1-NEXT: addq $-1, %rcx
+; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
+; AVX1-NEXT: addq $-1, %rdx
; AVX1-NEXT: movl $0, %r11d
; AVX1-NEXT: adcq $-1, %r11
-; AVX1-NEXT: addq $-1, %rbp
-; AVX1-NEXT: movl $0, %r10d
-; AVX1-NEXT: adcq $-1, %r10
-; AVX1-NEXT: addq $-1, %r15
+; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
+; AVX1-NEXT: addq $-1, %rax
+; AVX1-NEXT: movl $0, %ebx
+; AVX1-NEXT: adcq $-1, %rbx
+; AVX1-NEXT: addq $-1, %r12
+; AVX1-NEXT: movl $0, %r9d
+; AVX1-NEXT: adcq $-1, %r9
+; AVX1-NEXT: addq $-1, %r10
; AVX1-NEXT: movl $0, %r8d
; AVX1-NEXT: adcq $-1, %r8
-; AVX1-NEXT: addq $-1, %r9
-; AVX1-NEXT: movl $0, %edi
-; AVX1-NEXT: adcq $-1, %rdi
-; AVX1-NEXT: addq $-1, %rdx
-; AVX1-NEXT: movl $0, %eax
-; AVX1-NEXT: adcq $-1, %rax
-; AVX1-NEXT: shldq $63, %rdx, %rax
-; AVX1-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX1-NEXT: shldq $63, %r9, %rdi
-; AVX1-NEXT: shldq $63, %r15, %r8
-; AVX1-NEXT: shldq $63, %rbp, %r10
-; AVX1-NEXT: shldq $63, %rcx, %r11
-; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
-; AVX1-NEXT: shldq $63, %rdx, %rbx
+; AVX1-NEXT: addq $-1, %rdi
+; AVX1-NEXT: movl $0, %ecx
+; AVX1-NEXT: adcq $-1, %rcx
+; AVX1-NEXT: shldq $63, %rdi, %rcx
+; AVX1-NEXT: movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: shldq $63, %r10, %r8
+; AVX1-NEXT: shldq $63, %r12, %r9
+; AVX1-NEXT: shldq $63, %rax, %rbx
+; AVX1-NEXT: shldq $63, %rdx, %r11
; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
; AVX1-NEXT: shldq $63, %rdx, %r14
; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
; AVX1-NEXT: shldq $63, %rdx, %r13
-; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
-; AVX1-NEXT: shldq $63, %rdx, %r12
; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; AVX1-NEXT: shldq $63, %rax, %rsi
; AVX1-NEXT: movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
+; AVX1-NEXT: shldq $63, %rax, %r15
+; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
+; AVX1-NEXT: shldq $63, %rax, %rbp
; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; AVX1-NEXT: shldq $63, %rax, %rsi
-; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Reload
+; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; AVX1-NEXT: shldq $63, %rax, %r15
+; AVX1-NEXT: shldq $63, %rax, %rcx
+; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; AVX1-NEXT: shldq $63, %rcx, %rax
-; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
-; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; AVX1-NEXT: shldq $63, %rcx, %r9
-; AVX1-NEXT: movq (%rsp), %rcx # 8-byte Reload
-; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
-; AVX1-NEXT: shldq $63, %rdx, %rcx
+; AVX1-NEXT: shldq $63, %rax, %rdi
+; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Reload
+; AVX1-NEXT: movq (%rsp), %rax # 8-byte Reload
+; AVX1-NEXT: shldq $63, %rax, %r12
+; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
+; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
+; AVX1-NEXT: shldq $63, %rax, %r10
; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
-; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
-; AVX1-NEXT: shldq $63, %rbp, %rdx
-; AVX1-NEXT: vmovq %rdx, %xmm8
-; AVX1-NEXT: vmovq %rcx, %xmm0
-; AVX1-NEXT: vmovq %r9, %xmm1
-; AVX1-NEXT: vmovq %rax, %xmm11
-; AVX1-NEXT: vmovq %r15, %xmm2
+; AVX1-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
+; AVX1-NEXT: shldq $63, %rdx, %rax
+; AVX1-NEXT: vmovq %rax, %xmm8
+; AVX1-NEXT: vmovq %r10, %xmm0
+; AVX1-NEXT: vmovq %r12, %xmm1
+; AVX1-NEXT: vmovq %rdi, %xmm11
+; AVX1-NEXT: vmovq %rcx, %xmm2
; AVX1-NEXT: vmovq %rsi, %xmm13
-; AVX1-NEXT: vmovq {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 8-byte Folded Reload
-; AVX1-NEXT: # xmm14 = mem[0],zero
-; AVX1-NEXT: vmovq %r12, %xmm15
-; AVX1-NEXT: vmovq %r13, %xmm9
-; AVX1-NEXT: vmovq %r14, %xmm10
-; AVX1-NEXT: vmovq %rbx, %xmm12
+; AVX1-NEXT: vmovq %rbp, %xmm14
+; AVX1-NEXT: vmovq %r15, %xmm15
+; AVX1-NEXT: vmovq {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 8-byte Folded Reload
+; AVX1-NEXT: # xmm9 = mem[0],zero
+; AVX1-NEXT: vmovq %r13, %xmm10
+; AVX1-NEXT: vmovq %r14, %xmm12
; AVX1-NEXT: vmovq %r11, %xmm3
-; AVX1-NEXT: vmovq %r10, %xmm4
-; AVX1-NEXT: vmovq %r8, %xmm5
-; AVX1-NEXT: vmovq %rdi, %xmm6
+; AVX1-NEXT: vmovq %rbx, %xmm4
+; AVX1-NEXT: vmovq %r9, %xmm5
+; AVX1-NEXT: vmovq %r8, %xmm6
; AVX1-NEXT: vmovq {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 8-byte Folded Reload
; AVX1-NEXT: # xmm7 = mem[0],zero
; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm8 = xmm0[0],xmm8[0]
; AVX1-NEXT: vpshufb %xmm4, %xmm1, %xmm1
; AVX1-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX1-NEXT: vmovdqu %xmm0, (%rax)
-; AVX1-NEXT: addq $16, %rsp
+; AVX1-NEXT: addq $24, %rsp
; AVX1-NEXT: popq %rbx
; AVX1-NEXT: popq %r12
; AVX1-NEXT: popq %r13
; AVX2-NEXT: vpmovzxdq {{.*#+}} ymm3 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
; AVX2-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-NEXT: vpextrq $1, %xmm4, %rbx
-; AVX2-NEXT: vmovq %xmm4, %rdx
+; AVX2-NEXT: vmovq %xmm4, %rbp
; AVX2-NEXT: vpextrq $1, %xmm3, %rdi
; AVX2-NEXT: vmovq %xmm3, %rcx
; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm2
; AVX2-NEXT: vpmovzxdq {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm3
-; AVX2-NEXT: vpextrq $1, %xmm3, %r9
-; AVX2-NEXT: vmovq %xmm3, %r10
-; AVX2-NEXT: vpextrq $1, %xmm2, %r13
+; AVX2-NEXT: vpextrq $1, %xmm3, %rdx
+; AVX2-NEXT: vmovq %xmm3, %r9
+; AVX2-NEXT: vpextrq $1, %xmm2, %r11
; AVX2-NEXT: vmovq %xmm2, %r12
; AVX2-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; AVX2-NEXT: vpmovzxdq {{.*#+}} ymm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
; AVX2-NEXT: vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
; AVX2-NEXT: vpmovzxdq {{.*#+}} ymm3 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
; AVX2-NEXT: vextracti128 $1, %ymm3, %xmm4
-; AVX2-NEXT: vpextrq $1, %xmm4, %rbp
-; AVX2-NEXT: addq %rbx, %rbp
-; AVX2-NEXT: vmovq %xmm4, %rax
-; AVX2-NEXT: addq %rdx, %rax
-; AVX2-NEXT: movq %rax, %r11
-; AVX2-NEXT: vpextrq $1, %xmm3, %r8
-; AVX2-NEXT: addq %rdi, %r8
+; AVX2-NEXT: vpextrq $1, %xmm4, %rax
+; AVX2-NEXT: addq %rbx, %rax
+; AVX2-NEXT: movq %rax, %rbx
+; AVX2-NEXT: vmovq %xmm4, %r13
+; AVX2-NEXT: addq %rbp, %r13
+; AVX2-NEXT: vpextrq $1, %xmm3, %r10
+; AVX2-NEXT: addq %rdi, %r10
; AVX2-NEXT: vmovq %xmm3, %r14
; AVX2-NEXT: addq %rcx, %r14
; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm2
; AVX2-NEXT: vpmovzxdq {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX2-NEXT: vpextrq $1, %xmm3, %rax
-; AVX2-NEXT: addq %r9, %rax
-; AVX2-NEXT: movq %rax, %rbx
-; AVX2-NEXT: vmovq %xmm3, %rax
-; AVX2-NEXT: addq %r10, %rax
-; AVX2-NEXT: movq %rax, %r10
-; AVX2-NEXT: vpextrq $1, %xmm2, %rcx
-; AVX2-NEXT: addq %r13, %rcx
+; AVX2-NEXT: addq %rdx, %rax
+; AVX2-NEXT: movq %rax, %rcx
+; AVX2-NEXT: vmovq %xmm3, %r8
+; AVX2-NEXT: addq %r9, %r8
+; AVX2-NEXT: vpextrq $1, %xmm2, %rax
+; AVX2-NEXT: addq %r11, %rax
+; AVX2-NEXT: movq %rax, %r11
; AVX2-NEXT: vmovq %xmm2, %rax
; AVX2-NEXT: addq %r12, %rax
; AVX2-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX2-NEXT: vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm2
-; AVX2-NEXT: vpextrq $1, %xmm2, %r12
-; AVX2-NEXT: addq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Folded Reload
+; AVX2-NEXT: vpextrq $1, %xmm2, %rbp
+; AVX2-NEXT: addq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Folded Reload
; AVX2-NEXT: vmovq %xmm2, %r9
; AVX2-NEXT: addq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Folded Reload
; AVX2-NEXT: vpextrq $1, %xmm1, %rax
; AVX2-NEXT: vmovq %xmm1, %rdx
; AVX2-NEXT: vmovq %xmm0, %rsi
; AVX2-NEXT: addq %rdx, %rsi
-; AVX2-NEXT: addq $-1, %rbp
-; AVX2-NEXT: movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX2-NEXT: addq $-1, %rbx
+; AVX2-NEXT: movq %rbx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT: movl $0, %eax
; AVX2-NEXT: adcq $-1, %rax
; AVX2-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX2-NEXT: addq $-1, %r11
-; AVX2-NEXT: movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX2-NEXT: addq $-1, %r13
+; AVX2-NEXT: movq %r13, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT: movl $0, %eax
; AVX2-NEXT: adcq $-1, %rax
; AVX2-NEXT: movq %rax, (%rsp) # 8-byte Spill
-; AVX2-NEXT: addq $-1, %r8
-; AVX2-NEXT: movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX2-NEXT: addq $-1, %r10
+; AVX2-NEXT: movq %r10, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT: movl $0, %eax
; AVX2-NEXT: adcq $-1, %rax
; AVX2-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT: addq $-1, %r14
; AVX2-NEXT: movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX2-NEXT: movl $0, %ebp
-; AVX2-NEXT: adcq $-1, %rbp
-; AVX2-NEXT: addq $-1, %rbx
-; AVX2-NEXT: movq %rbx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX2-NEXT: movl $0, %r13d
+; AVX2-NEXT: adcq $-1, %r13
+; AVX2-NEXT: addq $-1, %rcx
+; AVX2-NEXT: movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT: movl $0, %eax
; AVX2-NEXT: adcq $-1, %rax
; AVX2-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX2-NEXT: addq $-1, %r10
-; AVX2-NEXT: movq %r10, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX2-NEXT: addq $-1, %r8
+; AVX2-NEXT: movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT: movl $0, %r15d
; AVX2-NEXT: adcq $-1, %r15
-; AVX2-NEXT: addq $-1, %rcx
-; AVX2-NEXT: movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX2-NEXT: addq $-1, %r11
+; AVX2-NEXT: movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT: movl $0, %ebx
; AVX2-NEXT: adcq $-1, %rbx
; AVX2-NEXT: addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
; AVX2-NEXT: adcq $-1, %rax
; AVX2-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT: addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; AVX2-NEXT: movl $0, %r13d
-; AVX2-NEXT: adcq $-1, %r13
+; AVX2-NEXT: movl $0, %r12d
+; AVX2-NEXT: adcq $-1, %r12
; AVX2-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
; AVX2-NEXT: addq $-1, %rcx
; AVX2-NEXT: movl $0, %r11d
; AVX2-NEXT: adcq $-1, %r11
-; AVX2-NEXT: addq $-1, %r12
+; AVX2-NEXT: addq $-1, %rbp
; AVX2-NEXT: movl $0, %r14d
; AVX2-NEXT: adcq $-1, %r14
; AVX2-NEXT: addq $-1, %r9
; AVX2-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX2-NEXT: shldq $63, %rdi, %rdx
; AVX2-NEXT: shldq $63, %r9, %r10
-; AVX2-NEXT: shldq $63, %r12, %r14
+; AVX2-NEXT: shldq $63, %rbp, %r14
; AVX2-NEXT: shldq $63, %rcx, %r11
; AVX2-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; AVX2-NEXT: shldq $63, %rcx, %r13
+; AVX2-NEXT: shldq $63, %rcx, %r12
; AVX2-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
; AVX2-NEXT: shldq $63, %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
; AVX2-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
; AVX2-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
; AVX2-NEXT: shldq $63, %rcx, %rax
; AVX2-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; AVX2-NEXT: shldq $63, %rcx, %rbp
-; AVX2-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Reload
+; AVX2-NEXT: shldq $63, %rcx, %r13
+; AVX2-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
; AVX2-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; AVX2-NEXT: shldq $63, %rcx, %r12
+; AVX2-NEXT: shldq $63, %rcx, %rbp
; AVX2-NEXT: movq (%rsp), %rdi # 8-byte Reload
; AVX2-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
; AVX2-NEXT: shldq $63, %rcx, %rdi
; AVX2-NEXT: shldq $63, %rcx, %rsi
; AVX2-NEXT: vmovq %rsi, %xmm8
; AVX2-NEXT: vmovq %rdi, %xmm9
-; AVX2-NEXT: vmovq %r12, %xmm10
-; AVX2-NEXT: vmovq %rbp, %xmm11
+; AVX2-NEXT: vmovq %rbp, %xmm10
+; AVX2-NEXT: vmovq %r13, %xmm11
; AVX2-NEXT: vmovq %rax, %xmm12
; AVX2-NEXT: vmovq %r15, %xmm13
; AVX2-NEXT: vmovq %rbx, %xmm14
; AVX2-NEXT: vmovq %r9, %xmm0
; AVX2-NEXT: vmovq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 8-byte Folded Reload
; AVX2-NEXT: # xmm1 = mem[0],zero
-; AVX2-NEXT: vmovq %r13, %xmm2
+; AVX2-NEXT: vmovq %r12, %xmm2
; AVX2-NEXT: vmovq %r11, %xmm3
; AVX2-NEXT: vmovq %r14, %xmm4
; AVX2-NEXT: vmovq %r10, %xmm5
; AVX512-NEXT: pushq %r13
; AVX512-NEXT: pushq %r12
; AVX512-NEXT: pushq %rbx
-; AVX512-NEXT: subq $16, %rsp
+; AVX512-NEXT: subq $24, %rsp
; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero
; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm1 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero
; AVX512-NEXT: vpmovzxwd {{.*#+}} ymm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
; AVX512-NEXT: vmovq %xmm3, %rsi
; AVX512-NEXT: vpmovzxdq {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
; AVX512-NEXT: vextracti128 $1, %ymm2, %xmm3
-; AVX512-NEXT: vpextrq $1, %xmm3, %rcx
-; AVX512-NEXT: vmovq %xmm3, %r10
+; AVX512-NEXT: vpextrq $1, %xmm3, %rdx
+; AVX512-NEXT: vmovq %xmm3, %r8
; AVX512-NEXT: vpextrq $1, %xmm2, %r13
; AVX512-NEXT: vmovq %xmm2, %r12
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX512-NEXT: vpmovzxdq {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
; AVX512-NEXT: vextracti128 $1, %ymm2, %xmm3
-; AVX512-NEXT: vpextrq $1, %xmm3, %rdx
+; AVX512-NEXT: vpextrq $1, %xmm3, %r15
; AVX512-NEXT: vmovq %xmm3, %r14
; AVX512-NEXT: vpextrq $1, %xmm2, %r9
; AVX512-NEXT: vmovq %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
; AVX512-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX512-NEXT: vpmovzxdq {{.*#+}} ymm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero
; AVX512-NEXT: vextracti128 $1, %ymm3, %xmm4
-; AVX512-NEXT: vpextrq $1, %xmm4, %r11
-; AVX512-NEXT: addq %rbx, %r11
+; AVX512-NEXT: vpextrq $1, %xmm4, %rax
+; AVX512-NEXT: addq %rbx, %rax
+; AVX512-NEXT: movq %rax, %rbx
; AVX512-NEXT: vmovq %xmm4, %rax
; AVX512-NEXT: addq %rbp, %rax
-; AVX512-NEXT: movq %rax, %rbx
-; AVX512-NEXT: vpextrq $1, %xmm3, %r8
-; AVX512-NEXT: addq %rdi, %r8
-; AVX512-NEXT: vmovq %xmm3, %r15
-; AVX512-NEXT: addq %rsi, %r15
+; AVX512-NEXT: movq %rax, %rbp
+; AVX512-NEXT: vpextrq $1, %xmm3, %rax
+; AVX512-NEXT: addq %rdi, %rax
+; AVX512-NEXT: movq %rax, %rdi
+; AVX512-NEXT: vmovq %xmm3, %r10
+; AVX512-NEXT: addq %rsi, %r10
; AVX512-NEXT: vpmovzxdq {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
; AVX512-NEXT: vextracti128 $1, %ymm2, %xmm3
-; AVX512-NEXT: vpextrq $1, %xmm3, %rdi
-; AVX512-NEXT: addq %rcx, %rdi
+; AVX512-NEXT: vpextrq $1, %xmm3, %rcx
+; AVX512-NEXT: addq %rdx, %rcx
; AVX512-NEXT: vmovq %xmm3, %rax
-; AVX512-NEXT: addq %r10, %rax
-; AVX512-NEXT: movq %rax, %r10
+; AVX512-NEXT: addq %r8, %rax
+; AVX512-NEXT: movq %rax, %r8
; AVX512-NEXT: vpextrq $1, %xmm2, %rsi
; AVX512-NEXT: addq %r13, %rsi
-; AVX512-NEXT: vmovq %xmm2, %rax
-; AVX512-NEXT: addq %r12, %rax
-; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512-NEXT: vmovq %xmm2, %r11
+; AVX512-NEXT: addq %r12, %r11
; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm1
; AVX512-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm2
; AVX512-NEXT: vpmovzxdq {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
; AVX512-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX512-NEXT: vpextrq $1, %xmm3, %rax
-; AVX512-NEXT: addq %rdx, %rax
+; AVX512-NEXT: addq %r15, %rax
; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512-NEXT: vmovq %xmm3, %rax
; AVX512-NEXT: addq %r14, %rax
; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512-NEXT: vpmovzxdq {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX512-NEXT: vpextrq $1, %xmm2, %rbp
-; AVX512-NEXT: addq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Folded Reload
+; AVX512-NEXT: vpextrq $1, %xmm2, %rax
+; AVX512-NEXT: addq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Folded Reload
+; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512-NEXT: vmovq %xmm2, %r14
; AVX512-NEXT: addq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Folded Reload
; AVX512-NEXT: vpextrq $1, %xmm0, %rax
; AVX512-NEXT: vpextrq $1, %xmm1, %r9
; AVX512-NEXT: addq %rax, %r9
-; AVX512-NEXT: vmovq %xmm0, %rcx
+; AVX512-NEXT: vmovq %xmm0, %rax
; AVX512-NEXT: vmovq %xmm1, %rdx
-; AVX512-NEXT: addq %rcx, %rdx
-; AVX512-NEXT: addq $-1, %r11
-; AVX512-NEXT: movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX512-NEXT: movl $0, %eax
-; AVX512-NEXT: adcq $-1, %rax
-; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512-NEXT: addq %rax, %rdx
; AVX512-NEXT: addq $-1, %rbx
; AVX512-NEXT: movq %rbx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512-NEXT: movl $0, %eax
; AVX512-NEXT: adcq $-1, %rax
-; AVX512-NEXT: movq %rax, (%rsp) # 8-byte Spill
-; AVX512-NEXT: addq $-1, %r8
-; AVX512-NEXT: movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX512-NEXT: movl $0, %eax
-; AVX512-NEXT: adcq $-1, %rax
; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX512-NEXT: addq $-1, %r15
-; AVX512-NEXT: movq %r15, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512-NEXT: addq $-1, %rbp
+; AVX512-NEXT: movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512-NEXT: movl $0, %eax
; AVX512-NEXT: adcq $-1, %rax
; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512-NEXT: movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512-NEXT: movl $0, %eax
; AVX512-NEXT: adcq $-1, %rax
-; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512-NEXT: movq %rax, (%rsp) # 8-byte Spill
; AVX512-NEXT: addq $-1, %r10
; AVX512-NEXT: movq %r10, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512-NEXT: movl $0, %eax
; AVX512-NEXT: adcq $-1, %rax
-; AVX512-NEXT: movq %rax, %rcx
+; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512-NEXT: addq $-1, %rcx
+; AVX512-NEXT: movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512-NEXT: movl $0, %eax
+; AVX512-NEXT: adcq $-1, %rax
+; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512-NEXT: addq $-1, %r8
+; AVX512-NEXT: movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512-NEXT: movl $0, %eax
+; AVX512-NEXT: adcq $-1, %rax
+; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512-NEXT: addq $-1, %rsi
; AVX512-NEXT: movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX512-NEXT: movl $0, %r12d
-; AVX512-NEXT: adcq $-1, %r12
-; AVX512-NEXT: addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; AVX512-NEXT: movl $0, %ebx
-; AVX512-NEXT: adcq $-1, %rbx
-; AVX512-NEXT: addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
; AVX512-NEXT: movl $0, %r13d
; AVX512-NEXT: adcq $-1, %r13
-; AVX512-NEXT: addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
+; AVX512-NEXT: addq $-1, %r11
+; AVX512-NEXT: movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512-NEXT: movl $0, %r15d
; AVX512-NEXT: adcq $-1, %r15
; AVX512-NEXT: addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; AVX512-NEXT: movl $0, %r11d
-; AVX512-NEXT: adcq $-1, %r11
+; AVX512-NEXT: movl $0, %eax
+; AVX512-NEXT: adcq $-1, %rax
+; AVX512-NEXT: movq %rax, %rsi
; AVX512-NEXT: addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; AVX512-NEXT: movl $0, %r8d
-; AVX512-NEXT: adcq $-1, %r8
+; AVX512-NEXT: movl $0, %r12d
+; AVX512-NEXT: adcq $-1, %r12
+; AVX512-NEXT: addq $-1, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
+; AVX512-NEXT: movl $0, %ebx
+; AVX512-NEXT: adcq $-1, %rbx
+; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
; AVX512-NEXT: addq $-1, %rbp
+; AVX512-NEXT: movl $0, %r11d
+; AVX512-NEXT: adcq $-1, %r11
+; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
+; AVX512-NEXT: addq $-1, %rax
; AVX512-NEXT: movl $0, %r10d
; AVX512-NEXT: adcq $-1, %r10
; AVX512-NEXT: addq $-1, %r14
+; AVX512-NEXT: movl $0, %r8d
+; AVX512-NEXT: adcq $-1, %r8
+; AVX512-NEXT: addq $-1, %r9
; AVX512-NEXT: movl $0, %edi
; AVX512-NEXT: adcq $-1, %rdi
-; AVX512-NEXT: addq $-1, %r9
-; AVX512-NEXT: movl $0, %esi
-; AVX512-NEXT: adcq $-1, %rsi
; AVX512-NEXT: addq $-1, %rdx
-; AVX512-NEXT: movl $0, %eax
-; AVX512-NEXT: adcq $-1, %rax
-; AVX512-NEXT: shldq $63, %rdx, %rax
-; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; AVX512-NEXT: shldq $63, %r9, %rsi
-; AVX512-NEXT: shldq $63, %r14, %rdi
-; AVX512-NEXT: shldq $63, %rbp, %r10
-; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
-; AVX512-NEXT: shldq $63, %rdx, %r8
+; AVX512-NEXT: movl $0, %ecx
+; AVX512-NEXT: adcq $-1, %rcx
+; AVX512-NEXT: shldq $63, %rdx, %rcx
+; AVX512-NEXT: movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512-NEXT: shldq $63, %r9, %rdi
+; AVX512-NEXT: shldq $63, %r14, %r8
+; AVX512-NEXT: shldq $63, %rax, %r10
+; AVX512-NEXT: shldq $63, %rbp, %r11
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
-; AVX512-NEXT: shldq $63, %rdx, %r11
+; AVX512-NEXT: shldq $63, %rdx, %rbx
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
-; AVX512-NEXT: shldq $63, %rdx, %r15
+; AVX512-NEXT: shldq $63, %rdx, %r12
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
-; AVX512-NEXT: shldq $63, %rdx, %r13
+; AVX512-NEXT: shldq $63, %rdx, %rsi
+; AVX512-NEXT: movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; AVX512-NEXT: shldq $63, %rax, %rbx
+; AVX512-NEXT: shldq $63, %rax, %r15
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; AVX512-NEXT: shldq $63, %rax, %r12
+; AVX512-NEXT: shldq $63, %rax, %r13
+; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; AVX512-NEXT: shldq $63, %rax, %rcx
-; AVX512-NEXT: movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; AVX512-NEXT: shldq $63, %rax, %rsi
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; AVX512-NEXT: shldq $63, %rax, %rcx
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
; AVX512-NEXT: shldq $63, %rdx, %rax
-; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Reload
+; AVX512-NEXT: movq (%rsp), %r14 # 8-byte Reload
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
; AVX512-NEXT: shldq $63, %rdx, %r14
-; AVX512-NEXT: movq (%rsp), %r9 # 8-byte Reload
+; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
; AVX512-NEXT: shldq $63, %rdx, %r9
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
-; AVX512-NEXT: shldq $63, %rbp, %rdx
-; AVX512-NEXT: vmovq %rdx, %xmm8
+; AVX512-NEXT: shldq $63, %rdx, %rbp
+; AVX512-NEXT: vmovq %rbp, %xmm8
; AVX512-NEXT: vmovq %r9, %xmm9
; AVX512-NEXT: vmovq %r14, %xmm10
; AVX512-NEXT: vmovq %rax, %xmm11
; AVX512-NEXT: vmovq %rcx, %xmm12
-; AVX512-NEXT: vmovq {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 8-byte Folded Reload
-; AVX512-NEXT: # xmm13 = mem[0],zero
-; AVX512-NEXT: vmovq %r12, %xmm14
-; AVX512-NEXT: vmovq %rbx, %xmm15
-; AVX512-NEXT: vmovq %r13, %xmm0
-; AVX512-NEXT: vmovq %r15, %xmm1
-; AVX512-NEXT: vmovq %r11, %xmm2
-; AVX512-NEXT: vmovq %r8, %xmm3
+; AVX512-NEXT: vmovq %rsi, %xmm13
+; AVX512-NEXT: vmovq %r13, %xmm14
+; AVX512-NEXT: vmovq %r15, %xmm15
+; AVX512-NEXT: vmovq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 8-byte Folded Reload
+; AVX512-NEXT: # xmm0 = mem[0],zero
+; AVX512-NEXT: vmovq %r12, %xmm1
+; AVX512-NEXT: vmovq %rbx, %xmm2
+; AVX512-NEXT: vmovq %r11, %xmm3
; AVX512-NEXT: vmovq %r10, %xmm4
-; AVX512-NEXT: vmovq %rdi, %xmm5
-; AVX512-NEXT: vmovq %rsi, %xmm6
+; AVX512-NEXT: vmovq %r8, %xmm5
+; AVX512-NEXT: vmovq %rdi, %xmm6
; AVX512-NEXT: vmovq {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 8-byte Folded Reload
; AVX512-NEXT: # xmm7 = mem[0],zero
; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm8 = xmm9[0],xmm8[0]
; AVX512-NEXT: vpmovqd %zmm0, %ymm0
; AVX512-NEXT: vinserti64x4 $1, %ymm0, %zmm2, %zmm0
; AVX512-NEXT: vpmovdb %zmm0, (%rax)
-; AVX512-NEXT: addq $16, %rsp
+; AVX512-NEXT: addq $24, %rsp
; AVX512-NEXT: popq %rbx
; AVX512-NEXT: popq %r12
; AVX512-NEXT: popq %r13
; ILP-NEXT: pushq %rbx
; ILP-NEXT: movq %rcx, %r9
; ILP-NEXT: movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; ILP-NEXT: xorl %eax, %eax
; ILP-NEXT: addq $1, %rsi
; ILP-NEXT: adcq $0, %rdx
; ILP-NEXT: adcq $0, %r9
; ILP-NEXT: adcq $0, %r8
; ILP-NEXT: leal 1(%rsi,%rsi), %edi
; ILP-NEXT: movl $1, %ebp
-; ILP-NEXT: xorl %eax, %eax
-; ILP-NEXT: xorl %r11d, %r11d
+; ILP-NEXT: xorl %r14d, %r14d
; ILP-NEXT: movl %edi, %ecx
-; ILP-NEXT: shldq %cl, %rbp, %r11
-; ILP-NEXT: movl $1, %r14d
-; ILP-NEXT: shlq %cl, %r14
+; ILP-NEXT: shldq %cl, %rbp, %r14
+; ILP-NEXT: movl $1, %r11d
+; ILP-NEXT: shlq %cl, %r11
; ILP-NEXT: movb $-128, %r10b
; ILP-NEXT: subb %dil, %r10b
; ILP-NEXT: movq %r9, %r13
; ILP-NEXT: xorl %r15d, %r15d
; ILP-NEXT: movl %edi, %ecx
; ILP-NEXT: shldq %cl, %r15, %r15
-; ILP-NEXT: movq %rsi, %rbp
-; ILP-NEXT: shrdq %cl, %rdx, %rbp
+; ILP-NEXT: movq %rsi, %rbx
+; ILP-NEXT: shrdq %cl, %rdx, %rbx
; ILP-NEXT: shrq %cl, %rdx
; ILP-NEXT: addb $-128, %cl
; ILP-NEXT: shrdq %cl, %r8, %r9
; ILP-NEXT: testb $64, %dil
-; ILP-NEXT: cmovneq %r14, %r11
-; ILP-NEXT: cmoveq %rbp, %rdx
+; ILP-NEXT: cmovneq %r11, %r14
+; ILP-NEXT: cmoveq %rbx, %rdx
; ILP-NEXT: cmovneq %rax, %r15
-; ILP-NEXT: cmovneq %rax, %r14
+; ILP-NEXT: cmovneq %rax, %r11
; ILP-NEXT: testb $64, %r10b
; ILP-NEXT: cmovneq %rax, %r12
; ILP-NEXT: cmovneq %rax, %r13
-; ILP-NEXT: movl $1, %ebp
-; ILP-NEXT: shlq %cl, %rbp
+; ILP-NEXT: movl $1, %ebx
+; ILP-NEXT: shlq %cl, %rbx
; ILP-NEXT: orl %edx, %r13d
; ILP-NEXT: xorl %edx, %edx
-; ILP-NEXT: movl $1, %ebx
-; ILP-NEXT: shldq %cl, %rbx, %rdx
+; ILP-NEXT: movl $1, %ebp
+; ILP-NEXT: shldq %cl, %rbp, %rdx
; ILP-NEXT: shrq %cl, %r8
; ILP-NEXT: testb $64, %cl
; ILP-NEXT: cmoveq %r9, %r8
-; ILP-NEXT: cmovneq %rbp, %rdx
-; ILP-NEXT: cmovneq %rax, %rbp
+; ILP-NEXT: cmovneq %rbx, %rdx
+; ILP-NEXT: cmovneq %rax, %rbx
; ILP-NEXT: testb %dil, %dil
-; ILP-NEXT: cmovsq %rax, %r11
; ILP-NEXT: cmovsq %rax, %r14
+; ILP-NEXT: cmovsq %rax, %r11
; ILP-NEXT: jns .LBB0_2
; ILP-NEXT: # %bb.1:
; ILP-NEXT: movl %r8d, %r13d
; ILP-NEXT: # %bb.3:
; ILP-NEXT: movl %r13d, %esi
; ILP-NEXT: .LBB0_4:
-; ILP-NEXT: cmovnsq %r12, %rbp
-; ILP-NEXT: cmoveq %rax, %rbp
+; ILP-NEXT: cmovnsq %r12, %rbx
+; ILP-NEXT: cmoveq %rax, %rbx
; ILP-NEXT: cmovnsq %r15, %rdx
; ILP-NEXT: cmoveq %rax, %rdx
; ILP-NEXT: testb $1, %sil
; ILP-NEXT: cmovneq %rax, %rdx
; ILP-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; ILP-NEXT: movq %rdx, 24(%rax)
-; ILP-NEXT: cmovneq %rax, %rbp
-; ILP-NEXT: movq %rbp, 16(%rax)
-; ILP-NEXT: cmovneq %rax, %r11
-; ILP-NEXT: movq %r11, 8(%rax)
+; ILP-NEXT: cmovneq %rax, %rbx
+; ILP-NEXT: movq %rbx, 16(%rax)
; ILP-NEXT: cmovneq %rax, %r14
-; ILP-NEXT: movq %r14, (%rax)
+; ILP-NEXT: movq %r14, 8(%rax)
+; ILP-NEXT: cmovneq %rax, %r11
+; ILP-NEXT: movq %r11, (%rax)
; ILP-NEXT: popq %rbx
; ILP-NEXT: popq %r12
; ILP-NEXT: popq %r13
;
; HYBRID-LABEL: test1:
; HYBRID: # %bb.0:
+; HYBRID-NEXT: pushq %rbp
; HYBRID-NEXT: pushq %r15
; HYBRID-NEXT: pushq %r14
; HYBRID-NEXT: pushq %r13
; HYBRID-NEXT: adcq $0, %rdx
; HYBRID-NEXT: adcq $0, %r9
; HYBRID-NEXT: adcq $0, %r8
+; HYBRID-NEXT: xorl %r10d, %r10d
; HYBRID-NEXT: leal 1(%rsi,%rsi), %edi
; HYBRID-NEXT: xorl %r14d, %r14d
-; HYBRID-NEXT: xorl %r15d, %r15d
; HYBRID-NEXT: movl %edi, %ecx
-; HYBRID-NEXT: shldq %cl, %r15, %r15
+; HYBRID-NEXT: shldq %cl, %r14, %r14
; HYBRID-NEXT: testb $64, %dil
-; HYBRID-NEXT: cmovneq %r14, %r15
-; HYBRID-NEXT: movl $1, %r11d
+; HYBRID-NEXT: cmovneq %r10, %r14
+; HYBRID-NEXT: movl $1, %ebp
; HYBRID-NEXT: movl $1, %r12d
; HYBRID-NEXT: shlq %cl, %r12
; HYBRID-NEXT: testb $64, %dil
-; HYBRID-NEXT: movq %r12, %r10
-; HYBRID-NEXT: cmovneq %r14, %r10
+; HYBRID-NEXT: movq %r12, %r11
+; HYBRID-NEXT: cmovneq %r10, %r11
; HYBRID-NEXT: movq %rsi, %rbx
; HYBRID-NEXT: shrdq %cl, %rdx, %rbx
; HYBRID-NEXT: shrq %cl, %rdx
; HYBRID-NEXT: testb $64, %dil
; HYBRID-NEXT: cmoveq %rbx, %rdx
-; HYBRID-NEXT: xorl %r13d, %r13d
-; HYBRID-NEXT: shldq %cl, %r11, %r13
+; HYBRID-NEXT: xorl %r15d, %r15d
+; HYBRID-NEXT: shldq %cl, %rbp, %r15
; HYBRID-NEXT: testb $64, %dil
-; HYBRID-NEXT: cmovneq %r12, %r13
+; HYBRID-NEXT: cmovneq %r12, %r15
; HYBRID-NEXT: movb $-128, %cl
; HYBRID-NEXT: subb %dil, %cl
-; HYBRID-NEXT: movq %r9, %rbx
-; HYBRID-NEXT: shlq %cl, %rbx
+; HYBRID-NEXT: movq %r9, %r13
+; HYBRID-NEXT: shlq %cl, %r13
; HYBRID-NEXT: movl $1, %r12d
-; HYBRID-NEXT: shrdq %cl, %r14, %r12
+; HYBRID-NEXT: shrdq %cl, %r10, %r12
; HYBRID-NEXT: testb $64, %cl
-; HYBRID-NEXT: cmovneq %r14, %r12
-; HYBRID-NEXT: cmovneq %r14, %rbx
-; HYBRID-NEXT: orl %edx, %ebx
+; HYBRID-NEXT: cmovneq %r10, %r12
+; HYBRID-NEXT: cmovneq %r10, %r13
+; HYBRID-NEXT: orl %edx, %r13d
; HYBRID-NEXT: movl %edi, %ecx
; HYBRID-NEXT: addb $-128, %cl
; HYBRID-NEXT: shrdq %cl, %r8, %r9
; HYBRID-NEXT: shrq %cl, %r8
; HYBRID-NEXT: xorl %edx, %edx
-; HYBRID-NEXT: shldq %cl, %r11, %rdx
-; HYBRID-NEXT: shlq %cl, %r11
+; HYBRID-NEXT: shldq %cl, %rbp, %rdx
+; HYBRID-NEXT: shlq %cl, %rbp
; HYBRID-NEXT: testb $64, %cl
-; HYBRID-NEXT: cmovneq %r11, %rdx
+; HYBRID-NEXT: cmovneq %rbp, %rdx
; HYBRID-NEXT: cmoveq %r9, %r8
-; HYBRID-NEXT: cmovneq %r14, %r11
+; HYBRID-NEXT: cmovneq %r10, %rbp
; HYBRID-NEXT: testb %dil, %dil
; HYBRID-NEXT: jns .LBB0_2
; HYBRID-NEXT: # %bb.1:
-; HYBRID-NEXT: movl %r8d, %ebx
+; HYBRID-NEXT: movl %r8d, %r13d
; HYBRID-NEXT: .LBB0_2:
; HYBRID-NEXT: je .LBB0_4
; HYBRID-NEXT: # %bb.3:
-; HYBRID-NEXT: movl %ebx, %esi
+; HYBRID-NEXT: movl %r13d, %esi
; HYBRID-NEXT: .LBB0_4:
-; HYBRID-NEXT: cmovsq %r14, %r13
-; HYBRID-NEXT: cmovnsq %r12, %r11
-; HYBRID-NEXT: cmoveq %r14, %r11
-; HYBRID-NEXT: cmovnsq %r15, %rdx
-; HYBRID-NEXT: cmoveq %r14, %rdx
-; HYBRID-NEXT: cmovsq %r14, %r10
+; HYBRID-NEXT: cmovsq %r10, %r15
+; HYBRID-NEXT: cmovnsq %r12, %rbp
+; HYBRID-NEXT: cmoveq %r10, %rbp
+; HYBRID-NEXT: cmovnsq %r14, %rdx
+; HYBRID-NEXT: cmoveq %r10, %rdx
+; HYBRID-NEXT: cmovsq %r10, %r11
; HYBRID-NEXT: testb $1, %sil
; HYBRID-NEXT: cmovneq %rax, %rdx
; HYBRID-NEXT: movq %rdx, 24(%rax)
+; HYBRID-NEXT: cmovneq %rax, %rbp
+; HYBRID-NEXT: movq %rbp, 16(%rax)
+; HYBRID-NEXT: cmovneq %rax, %r15
+; HYBRID-NEXT: movq %r15, 8(%rax)
; HYBRID-NEXT: cmovneq %rax, %r11
-; HYBRID-NEXT: movq %r11, 16(%rax)
-; HYBRID-NEXT: cmovneq %rax, %r13
-; HYBRID-NEXT: movq %r13, 8(%rax)
-; HYBRID-NEXT: cmovneq %rax, %r10
-; HYBRID-NEXT: movq %r10, (%rax)
+; HYBRID-NEXT: movq %r11, (%rax)
; HYBRID-NEXT: popq %rbx
; HYBRID-NEXT: popq %r12
; HYBRID-NEXT: popq %r13
; HYBRID-NEXT: popq %r14
; HYBRID-NEXT: popq %r15
+; HYBRID-NEXT: popq %rbp
; HYBRID-NEXT: retq
;
; BURR-LABEL: test1:
; BURR: # %bb.0:
+; BURR-NEXT: pushq %rbp
; BURR-NEXT: pushq %r15
; BURR-NEXT: pushq %r14
; BURR-NEXT: pushq %r13
; BURR-NEXT: adcq $0, %rdx
; BURR-NEXT: adcq $0, %r9
; BURR-NEXT: adcq $0, %r8
+; BURR-NEXT: xorl %r10d, %r10d
; BURR-NEXT: leal 1(%rsi,%rsi), %edi
; BURR-NEXT: xorl %r14d, %r14d
-; BURR-NEXT: xorl %r15d, %r15d
; BURR-NEXT: movl %edi, %ecx
-; BURR-NEXT: shldq %cl, %r15, %r15
+; BURR-NEXT: shldq %cl, %r14, %r14
; BURR-NEXT: testb $64, %dil
-; BURR-NEXT: cmovneq %r14, %r15
-; BURR-NEXT: movl $1, %r11d
+; BURR-NEXT: cmovneq %r10, %r14
+; BURR-NEXT: movl $1, %ebp
; BURR-NEXT: movl $1, %r12d
; BURR-NEXT: shlq %cl, %r12
; BURR-NEXT: testb $64, %dil
-; BURR-NEXT: movq %r12, %r10
-; BURR-NEXT: cmovneq %r14, %r10
+; BURR-NEXT: movq %r12, %r11
+; BURR-NEXT: cmovneq %r10, %r11
; BURR-NEXT: movq %rsi, %rbx
; BURR-NEXT: shrdq %cl, %rdx, %rbx
; BURR-NEXT: shrq %cl, %rdx
; BURR-NEXT: testb $64, %dil
; BURR-NEXT: cmoveq %rbx, %rdx
-; BURR-NEXT: xorl %r13d, %r13d
-; BURR-NEXT: shldq %cl, %r11, %r13
+; BURR-NEXT: xorl %r15d, %r15d
+; BURR-NEXT: shldq %cl, %rbp, %r15
; BURR-NEXT: testb $64, %dil
-; BURR-NEXT: cmovneq %r12, %r13
+; BURR-NEXT: cmovneq %r12, %r15
; BURR-NEXT: movb $-128, %cl
; BURR-NEXT: subb %dil, %cl
-; BURR-NEXT: movq %r9, %rbx
-; BURR-NEXT: shlq %cl, %rbx
+; BURR-NEXT: movq %r9, %r13
+; BURR-NEXT: shlq %cl, %r13
; BURR-NEXT: movl $1, %r12d
-; BURR-NEXT: shrdq %cl, %r14, %r12
+; BURR-NEXT: shrdq %cl, %r10, %r12
; BURR-NEXT: testb $64, %cl
-; BURR-NEXT: cmovneq %r14, %r12
-; BURR-NEXT: cmovneq %r14, %rbx
-; BURR-NEXT: orl %edx, %ebx
+; BURR-NEXT: cmovneq %r10, %r12
+; BURR-NEXT: cmovneq %r10, %r13
+; BURR-NEXT: orl %edx, %r13d
; BURR-NEXT: movl %edi, %ecx
; BURR-NEXT: addb $-128, %cl
; BURR-NEXT: shrdq %cl, %r8, %r9
; BURR-NEXT: xorl %edx, %edx
-; BURR-NEXT: shldq %cl, %r11, %rdx
+; BURR-NEXT: shldq %cl, %rbp, %rdx
; BURR-NEXT: shrq %cl, %r8
-; BURR-NEXT: shlq %cl, %r11
+; BURR-NEXT: shlq %cl, %rbp
; BURR-NEXT: testb $64, %cl
-; BURR-NEXT: cmovneq %r11, %rdx
+; BURR-NEXT: cmovneq %rbp, %rdx
; BURR-NEXT: cmoveq %r9, %r8
-; BURR-NEXT: cmovneq %r14, %r11
+; BURR-NEXT: cmovneq %r10, %rbp
; BURR-NEXT: testb %dil, %dil
; BURR-NEXT: jns .LBB0_2
; BURR-NEXT: # %bb.1:
-; BURR-NEXT: movl %r8d, %ebx
+; BURR-NEXT: movl %r8d, %r13d
; BURR-NEXT: .LBB0_2:
; BURR-NEXT: je .LBB0_4
; BURR-NEXT: # %bb.3:
-; BURR-NEXT: movl %ebx, %esi
+; BURR-NEXT: movl %r13d, %esi
; BURR-NEXT: .LBB0_4:
-; BURR-NEXT: cmovsq %r14, %r13
-; BURR-NEXT: cmovnsq %r12, %r11
-; BURR-NEXT: cmoveq %r14, %r11
-; BURR-NEXT: cmovnsq %r15, %rdx
-; BURR-NEXT: cmoveq %r14, %rdx
-; BURR-NEXT: cmovsq %r14, %r10
+; BURR-NEXT: cmovsq %r10, %r15
+; BURR-NEXT: cmovnsq %r12, %rbp
+; BURR-NEXT: cmoveq %r10, %rbp
+; BURR-NEXT: cmovnsq %r14, %rdx
+; BURR-NEXT: cmoveq %r10, %rdx
+; BURR-NEXT: cmovsq %r10, %r11
; BURR-NEXT: testb $1, %sil
; BURR-NEXT: cmovneq %rax, %rdx
; BURR-NEXT: movq %rdx, 24(%rax)
+; BURR-NEXT: cmovneq %rax, %rbp
+; BURR-NEXT: movq %rbp, 16(%rax)
+; BURR-NEXT: cmovneq %rax, %r15
+; BURR-NEXT: movq %r15, 8(%rax)
; BURR-NEXT: cmovneq %rax, %r11
-; BURR-NEXT: movq %r11, 16(%rax)
-; BURR-NEXT: cmovneq %rax, %r13
-; BURR-NEXT: movq %r13, 8(%rax)
-; BURR-NEXT: cmovneq %rax, %r10
-; BURR-NEXT: movq %r10, (%rax)
+; BURR-NEXT: movq %r11, (%rax)
; BURR-NEXT: popq %rbx
; BURR-NEXT: popq %r12
; BURR-NEXT: popq %r13
; BURR-NEXT: popq %r14
; BURR-NEXT: popq %r15
+; BURR-NEXT: popq %rbp
; BURR-NEXT: retq
;
; SRC-LABEL: test1:
; SRC-NEXT: movl %r11d, %ecx
; SRC-NEXT: shrdq %cl, %rdx, %rbp
; SRC-NEXT: shrq %cl, %rdx
-; SRC-NEXT: movl $1, %edi
; SRC-NEXT: xorl %r15d, %r15d
+; SRC-NEXT: movl $1, %edi
; SRC-NEXT: xorl %r14d, %r14d
; SRC-NEXT: shldq %cl, %rdi, %r14
; SRC-NEXT: xorl %r13d, %r13d
; ILP-LABEL: test4:
; ILP: # %bb.0:
; ILP-NEXT: xorl %ecx, %ecx
+; ILP-NEXT: xorl %edx, %edx
; ILP-NEXT: addq $1, %rsi
-; ILP-NEXT: setb %cl
+; ILP-NEXT: setb %dl
; ILP-NEXT: movl $2, %eax
-; ILP-NEXT: xorl %edx, %edx
; ILP-NEXT: cmpq %rdi, %rsi
-; ILP-NEXT: sbbq $0, %rcx
-; ILP-NEXT: movl $0, %ecx
-; ILP-NEXT: sbbq $0, %rcx
; ILP-NEXT: sbbq $0, %rdx
+; ILP-NEXT: movl $0, %edx
+; ILP-NEXT: sbbq $0, %rdx
+; ILP-NEXT: sbbq $0, %rcx
; ILP-NEXT: setae %cl
; ILP-NEXT: movzbl %cl, %ecx
; ILP-NEXT: subq %rcx, %rax
; HYBRID-LABEL: test4:
; HYBRID: # %bb.0:
; HYBRID-NEXT: xorl %eax, %eax
-; HYBRID-NEXT: addq $1, %rsi
-; HYBRID-NEXT: setb %al
; HYBRID-NEXT: xorl %ecx, %ecx
+; HYBRID-NEXT: addq $1, %rsi
+; HYBRID-NEXT: setb %cl
; HYBRID-NEXT: cmpq %rdi, %rsi
-; HYBRID-NEXT: sbbq $0, %rax
-; HYBRID-NEXT: movl $0, %eax
-; HYBRID-NEXT: sbbq $0, %rax
; HYBRID-NEXT: sbbq $0, %rcx
+; HYBRID-NEXT: movl $0, %ecx
+; HYBRID-NEXT: sbbq $0, %rcx
+; HYBRID-NEXT: sbbq $0, %rax
; HYBRID-NEXT: setae %al
; HYBRID-NEXT: movzbl %al, %ecx
; HYBRID-NEXT: movl $2, %eax
; BURR-LABEL: test4:
; BURR: # %bb.0:
; BURR-NEXT: xorl %eax, %eax
-; BURR-NEXT: addq $1, %rsi
-; BURR-NEXT: setb %al
; BURR-NEXT: xorl %ecx, %ecx
+; BURR-NEXT: addq $1, %rsi
+; BURR-NEXT: setb %cl
; BURR-NEXT: cmpq %rdi, %rsi
-; BURR-NEXT: sbbq $0, %rax
-; BURR-NEXT: movl $0, %eax
-; BURR-NEXT: sbbq $0, %rax
; BURR-NEXT: sbbq $0, %rcx
+; BURR-NEXT: movl $0, %ecx
+; BURR-NEXT: sbbq $0, %rcx
+; BURR-NEXT: sbbq $0, %rax
; BURR-NEXT: setae %al
; BURR-NEXT: movzbl %al, %ecx
; BURR-NEXT: movl $2, %eax